Performance on AMD Bulldozer

[wanne]

Let me check if something else uses max render threads as a limit.

I didn't play around with the render threads much. The basic idea behind this was restricting factorio to less cores to get more cache hits. But if I restricting the use of the other cores anyway it makes no sense to have 8 rendering threads. This was the reason, why i changed it down.
I at first changed restricting strictly to one core what made things worse instead of better. Than I tried a little bit around. The main result is: Using 4 cores seems to work much better than using one or 8. This was strictly done on OS-Side. Not within factorio. – I just changed the rendering threads to the same value.

Since help is not what seem to be looking for I will close this.

This is not true. I would be really interested in answers to the original Question. What do you need to play huge maps. The only answer to that question came from Jap2.0 saying that he can play with a much slower CPU.

Everybody else just got his own topic: BlueTemplar insisting that the CPU is way below average also it would be ranked on place 25 of 100 of the most common CPUs in his own source. (A benchmarking software which has for sure an other user circle than the normal office-user.) And CrushedIce who started discuss about performance optimization and at last posila who seems to be much more interested in how you could optimize the game. I answered to these off-topic things. This does not mean, that I'm not interested in an answer to the original question. – Should I try to make a different tread which stays more on topic?

[wanne]

What's your UPS in this save? (you need to update to 0.18 experimental, if you haven't already): https://www.dropbox.com/s/hkqbnymgpl211 ... 8.zip?dl=0

Both FPS and UPS are around 38-39 without pinning and around 45 with pinning.

[azesmbog]

wanne, but what is the performance in the standard test?
https://factoriobox.1au.us/
if of course it’s not difficult to run?

[wanne]

Update: In benchmarking mode I get more (for the charityStream12):

Code: Select all

bin/x64/factorio --mod-directory /dev/null --benchmark saves/charityStream12_018.zip --benchmark-runs 2 --benchmark-sanitize
Performed 1000 updates in 21829.754 ms
  avg: 21.830 ms, min: 19.044 ms, max: 39.843 ms
  checksum: 3330664199
Performed 1000 updates in 22095.718 ms
  avg: 22.096 ms, min: 19.560 ms, max: 43.368 ms
  checksum: 3330664199

=> ~45UPS

Code: Select all

taskset -a 3 bin/x64/factorio --mod-directory /dev/null --benchmark saves/charityStream12_018.zip --benchmark-runs 2 --benchmark-sanitize
avg: 20.429 ms, min: 19.007 ms, max: 38.068 ms
  checksum: 3330664199
  Performed 1000 updates in 20582.085 ms
  avg: 20.582 ms, min: 19.329 ms, max: 38.430 ms
  checksum: 3330664199

=> ~49UPS

[wanne]

wanne, but what is the performance in the standard test?
https://factoriobox.1au.us/
if of course it’s not difficult to run?

With pinning

Code: Select all

  Performed 1000 updates in 24431.193 ms
  avg: 24.431 ms, min: 22.415 ms, max: 57.116 ms
  checksum: 398646746
  Performed 1000 updates in 24559.900 ms
  avg: 24.560 ms, min: 22.451 ms, max: 57.591 ms
  checksum: 398646746

=> 41UPS
without:

Code: Select all

  Performed 1000 updates in 27592.184 ms
  avg: 27.592 ms, min: 23.772 ms, max: 60.684 ms
  checksum: 398646746
  Performed 1000 updates in 27752.086 ms
  avg: 27.752 ms, min: 23.481 ms, max: 60.425 ms
  checksum: 398646746

=> 36UPS

So it does not make that much of a difference. But it seems that that difference is it that makes me slower of faster than the server.

[wanne]

Btw. is this a kind or an answer to the original question:
When I look at https://factoriobox.1au.us/results I can search for CPUs that are above 60UPS and I should be fine?

[azesmbog]

this is the answer to the question why the ancient four! core 8400 shows exactly the same 41 ups.
https://factoriobox.1au.us/result/e849f ... 83a427abe2
More convenient to compare

[Bilka]

When I look at https://factoriobox.1au.us/results I can search for CPUs that are above 60UPS and I should be fine?

The standard test map (cb0cd35aa6893dfdae2ce574d345d2c2) is a 10k spm map, so it is not "join smaller multiplayer sessions" which is what you ask for in your first post. Having a CPU that can run an (UPS optimized) 10k spm map at 60 UPS is massive overkill for small online sessions.

What you describe as small "Playercount was between 1 and three." should be possible to run on basically anything, assuming that you don't mean "3 people but 3k hours of playtime and a scenario script that is bad for performance", but "3 people and 50 hours of playtime playing 100% vanilla". As BlueTemplar says, a server like comfy's biter battles can run like sh*t with only 6 hours of playtime and 40-50 players due to the scenario scripts and the impact of the massive biter attacks they generate.

From personal experience: Some of the speedrunners have non-highend PC's, they usually cannot play biter battles to the end due to performance problems, same with 100% multiplayer speedruns which push the game directly to its limits. However, they never have any problems in our multiplayer speedruns, which is 8 people launching the rocket as fast as possible which does mean that the factory is not small.

All of these examples really just illustrate that it very much matters what servers you are joining - how big is the factory, are there any custom scripts, etc. So, please try to join some servers again and perhaps provide screenshots or saves so that we can know what you are comparing to. And then we can give recommendations based on that.

Your performance on the charitymap provided by posila is better than that of my 6 year old laptop, I get around 38-37 UPS. In the server list, joined "AssemblyStorm weekly", it seems to be doing a around 800spm base at 90+h playtime, a map that I'd consider medium to big. My laptop could barely keep up with the map, but that is already more than I expected from that poor AMD A10-5745M. That's below min specs, just as a note. Could you join that server (it's on 0.18.2) and see how things go for you?

[wanne]

Your performance on the charitymap provided by posila is better than that of my 6 year old laptop […] more than I expected from that poor AMD A10-5745M.

Should be. A 2013 low end Laptop CPU should not be faster than a high end 2014 Hich-End-Desktop-CPU.

Could you join that server (it's on 0.18.2) and see how things go for you?

No time at the moment. Maybe I test later. But so far it seems for me that in the end I just have to change hardware or skip playing multiplayer. (I'm not the speedrun guy I like these Bob/seablock... games were you have to think big and plan things right.)

his is the answer to the question why the ancient four! core 8400 shows exactly the same 41 ups.
https://factoriobox.1au.us/result/e849f ... 83a427abe2

So the $20 Intel CPU (PassMark Score 1120 single core/3161 overall) outperfomrs the $200 AMD CPU (PassMark Score 1537 single core/9046 overall) (Both amazon prices for new ones. Sure there are cheaper prices on other platforms.) without optimizations by 14%.
Or in other words: If you want to play factorio you don't need a faster CPU you need an Intel one.

[Bilka]

Or in other words: If you want to play factorio you don't need a faster CPU you need an Intel one.

My AMD Ryzen 9 3900x runs Factorio just fine, about 4 times as fast as my laptop You may want to keep in mind that the bulldozer CPU's are a few years old and don't represent the current lineup of AMD CPU's. Same thing for the Q8400 not representing the current Intel lineup. If you want to spend lots of money, Intel will serve you well for Factorio in the high-end, but in the middle ground you'll get more bang out of your buck with an AMD CPU like the 3600 or a 2600.

[azesmbog]

If you want to play factorio you don't need a faster CPU you need an Intel one.

See all the results.
https://factoriobox.1au.us/results
and draw conclusions yourself

Same thing for the Q8400 not representing the current Intel lineup.

I also have a Q8100, in the Core Duo 8400-8500.
Someday I will try the results on them.
There are still results on the "for all" I5-9400, but yesterday the results did not want to be added to the database.

[wanne]

If you want to spend lots of money, Intel will serve you well for Factorio in the high-end, but in the middle ground you'll get more bang out of your buck with an AMD CPU like the 3600 or a 2600.

When I read this, I doubt that: viewtopic.php?f=5&t=42165
Thinking about a Ryzen 7 2700X. Maybe a Ryzen 5 3600.
But it seems that the performance of it sucks too: https://factoriobox.1au.us/result/aedb3 ... 940110e9b0
Even my office PC (i5-6500) should produce better results. Here are the results of the slower i5-6400 https://factoriobox.1au.us/result/f3f0f ... da0f3feae6 (I have to try this later)
Seems that an i5-9400 or i5-9600 should achieve much better results (they are 3 generations newer) for the same price.

[azesmbog]

Here are the results of the slower i5-6400 https://factoriobox.1au.us/result/f3f0f ... da0f3feae6 (I have to try this later)
Seems that an i5-9400 or i5-9600 should achieve much better results (they are 3 generations newer) for the same price.

6400 - it’s not slow, it’s just old, but it’s overclocked quite seriously. Although there is still stock

6400_166.jpg (130.61 KiB) Viewed 5276 times

yesterday the site did not accept the results: ((

9400_3400_wn10.jpg (121.91 KiB) Viewed 5276 times

and these are the results of 9400, and it is not overclocked, but still. For comparison
results for 18.1

[Olacken]

When I read this, I doubt that: viewtopic.php?f=5&t=42165 ...

0.18 added a lot of optimisation so i'm not sure those older test are really still relevant and even if they were I wouldn't compare too much between two version that are that far apart more so when the last one show better performance

[BlueTemplar]

[...]

Since help is not what seem to be looking for I will close this.

This is not true. I would be really interested in answers to the original Question. What do you need to play huge maps. The only answer to that question came from Jap2.0 saying that he can play with a much slower CPU.

Everybody else just got his own topic: BlueTemplar insisting that the CPU is way below average also it would be ranked on place 25 of 100 of the most common CPUs in his own source. (A benchmarking software which has for sure an other user circle than the normal office-user.)
[...]

Uh, my post was about how you would "eventually" get dropped if your PC wasn't slightly faster than the server, due to how Factorio handles it by default,
how you had a CPU with worse single-threading than the worst (?) one in the Steam Hardware Survey most common category,*
EDIT : As a reminder, it's "Intel 3.3 Ghz to 3.69 Ghz", do any of your counter-examples fit that ?
and how I had a CPU similar to yours, and encountered similar issues.
How is this offtopic ??

*that single-threading score sadly doesn't seem to be completely relevant to Factorio, as Jap2.0 has shown...

(And if you have a better, at least somewhat statistics-backed, idea about what would be a "common" CPU, I'm all hears!)

But yeah, I should also have linked that benchmark thread too...

P.S.: And as a Bulldozer owner, shouldn't you be well-aware about its single-threaded shortcomings ?
For my part, I was tricked by Brad Wardell trying to push his own new game and claiming that soon(tm), even simulation-heavy games would be fully multi-threaded... (which isn't true for his own game, the Ashes of the Singularity simulation AFAIK is NOT multi-threaded !)
But I guess I also wanted this too much to be true ?

P.P.S.: And last but not least, you should know by now that *in practice*, devs generally do NOT support Bulldozers, as can be seen from the above :

[...]
I need to find where did our Bulldozer PC end up in the office so I can experiment ... Also, I might not get properly to it until May or so.
[...]

And which I suspect is the reason why my VR experience (which *does* multithread *extremely* well) is that 90% of programs work fine, and 10% are completely unusable. With supposedly the *same* minimum requirements !
But who can blame them : Bulldozers never had much of a market share among gamers, due a lot to this bad single-thread performance, even when they were the best that AMD could offer !

[wanne]

EDIT : As a reminder, it's "Intel 3.3 Ghz to 3.69 Ghz"

No, its not. The biggest Number is standing at INTEL CPU SPEEDS (OSX): 2.3 Ghz to 2.69 Ghz: 30%. But just searching for the biggest number is statistical bullshit. Even if you look only on Windows and only on Intel declaring the Category where only 6% are faster but 71% slower (in clock speed) to "average" is bullshit.
This is like declaring Sebastian Langeveld to a below than average road-rad-racer because he would be one of the last in the most common race: The Tour de France.

do any of your counter-examples fit that ?

There are many:
i5-660: 1393
i5-661: 1341
i7-990X: 1515
i3-560: 1352
But these don't fit in the category "most common" on PassMark. Filtering for Windows+Intel+Most common Clockspeeds+Most common on PassMark you get 4 CPUs. And yes all of these 4 are faster than mine but mine wouldn't fall in that category either. 95% of the CPUs used on steam don't fall in these definition of "common". There are a little bit more CPUs on the world than only 4. Grepping for most common Brand AND most common Clockspeeds AND Most common CPU instead of just most common CPU is totally made up to have a way to small totally non-representative set of CPUs.

And if you have a better, at least somewhat statistics-backed, idea about what would be a "common" CPU, I'm all hears!

Like I said just use the "most common 100 for 28. January 2020" category of PassMark there it would be on place 43 in single core speed.

[wanne]

And as a Bulldozer owner, shouldn't you be well-aware about its single-threaded shortcomings ?

Yes I know. But until now it was never a problem for factorio. Like I said I could play most of the maps even on my 9 year old Laptop. So I never assumed to get problems on the Desktop which runs many applications 10 times faster.
But I when I read it right this was just Luck, that that thing performed much then above what you would expect on factorio. And factorio was always a really demanding game which is really picky about which hardware it likes.

[Jap2.0]

If you want to play factorio you don't need a faster CPU you need an Intel one.

See all the results.
https://factoriobox.1au.us/results
and draw conclusions yourself

The conclusion I draw from that is as follows:

If you want to play factorio you need decent speed memory.

[wanne]

Here are the results for my office PC: https://factoriobox.1au.us/result/1042c ... d038ee07ba
It has overall better hardware stats than your i5-6400 system but is much slower.
So Overclocking brings you a lot.

yesterday the site did not accept the results:

I could upload the results for you. Most values are easy to guess or in your last post. But I don't think thats intended.
Btw. is there a contact to the guy who runs the site? There are a few mistakes in his linux-Script. (Basically it runs only on Ubuntu without the need while it is declared as generic.)

[azesmbog]

The conclusion I draw from that is as follows:

If you want to play factorio you need decent speed memory.

One-sided conclusion.
So far, the top AMD +3800 MHz memory with good timings is not lost at all by the top Intel with 2900 MHz memory. And if we talk about memory, then my database will not start at all with a memory capacity of 8 or 16 GB, it needs 32 and above. But how much higher is another question. and the comparison of processors (bulldozers) certainly does not apply )
and yes, 4400+ memory will not harm any application.

yesterday the site did not accept the results:

I could upload the results for you. Most values are easy to guess or in your last post. But I don't think thats intended.

I did tests at least 20 times yesterday. It’s not difficult for me)
One value only uploaded to the site, which is q8400.
Of these 20 times, always a stable result was 97,
but then I launched Windows 10 with the same hardware - a couple of times - the result was exactly 95, or rather three times, the first result was even smaller. This is a comparison of win7 vs win10. on Linux, probably the result is also slightly smaller, but not as much.